Letta's Claude Subconscious Gives Claude Code a Persistent Memory Across Sessions
Letta's open-source plugin runs a background agent that watches Claude Code sessions, remembers everything, and whispers guidance before your next prompt.
PRO- Letta released claude-subconscious, a Claude Code plugin adding a persistent background memory agent.
- Uses four hooks to watch transcripts and inject guidance via stdout, never touching CLAUDE.md.
- Default agent maintains eight memory blocks including user preferences, project context, and pending items.
- One agent brain is shared across all your projects via Letta's Conversations API.
- Free to try with bundled zai/glm-5 model on Letta Cloud; override with any major provider.
- Marked as a demo; the team steers production users to Letta Code.
Claude Code is stateless. Close the terminal, come back tomorrow, and the model has no idea what you built, why you picked that architecture, or that you always want explicit type annotations. Letta's new open-source plugin, claude-subconscious, attaches a second agent underneath every session that quietly accumulates memory and injects it back into your next prompt.
The repo has crossed 2.8k stars, ships under an MIT license, and installs as a Claude Code plugin. The creator describes it as a plugin that attaches a Letta agent's memory to Claude and relays the transcript between you and Claude back to a Letta agent, which can update its memory blocks to persist information across sessions.
The forgetful assistant problem
Even strong coding models feel forgetful. They lose track of preferences, architectural decisions, unresolved tasks, and the subtle patterns that make collaboration smoother over time. Claude Subconscious tries to solve that by giving Claude Code something closer to a persistent background mind.
The design goes beyond a scratchpad. By default, the Subconscious agent gets client-side tool access via the Letta Code SDK. Rather than being limited to memory operations, Sub can read your files, search the web, and explore your codebase while processing transcripts. In full mode it can even spawn sub-agents through the Task tool to run parallel research while you keep working.
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
